https://witnessedsentence.org/rss.xmlThe Witnessed Sentence — RegistryVerified cases of algorithmic harm2026-10-10T16:06:44.084191+00:00CC-BY 4.0 — https://creativecommons.org/licenses/by/4.0/https://witnessedsentence.org/registry/r-4572/YouTube machine-learning extremist-content flagging terminated Syrian human-rights documentation channels, 2017–20192026-10-10T16:06:44.084191+00:00YouTube used machine-learning automated flagging to remove extremist content and terminated thousands of Syrian channels that published videos of human rights violations, removing material usable as a public record of the conflict. A June 2019 report by EFF, Syrian Archive and WITNESS documented these removals alongside other over-broad takedowns of activism, satire and counter-speech. [United States]https://witnessedsentence.org/registry/r-3764/Intelligence-Led Policing youth risk-scoring and repeated home checks, Pasco County Sheriff's Office, Florida, United States, 2015–20242026-10-10T16:06:44.084191+00:00The Pasco County Sheriff's Office used a point-based rubric drawing on school grades, absences, minor infractions and victimisation records to label young people as likely future offenders. Deputies made more than 12,500 preemptive visits from 2015, which led to unrelated fines and arrests of family members. After 2020 journalism, a federal lawsuit ended in a 2024 settlement in which the Sheriff admitted constitutional violations. [United States, Florida (Pasco County)]https://witnessedsentence.org/registry/r-3703/Rikunabi job-offer-decline prediction scores sold to employers without consent, Recruit Career, Japan, 2018–20192026-10-10T16:06:44.084191+00:00Between March 2018 and February 2019, Recruit Career used Rikunabi job-site browsing data and a machine-learning model to score how likely individual student applicants were to decline job offers, selling the scores to 35 client companies. Japan's Personal Information Protection Commission found personal data was provided without consent, issued recommendations, and the service was discontinued in 2019. [Japan]https://witnessedsentence.org/registry/r-3573/ShotSpotter gunshot alerts used to justify investigatory stops and frisks, Chicago Police Department, Chicago, 2018–20212026-10-10T16:06:44.084191+00:00Chicago's Office of Inspector General analysed more than 50,000 ShotSpotter alert and dispatch records from the Chicago Police Department. It found that 9% of alerts with a reported disposition indicated a gun-related offence, and that officers cited aggregate ShotSpotter alert frequency in an area as partial grounds for investigatory stops and pat-down frisks of civilians. [United States, Chicago, Illinois]https://witnessedsentence.org/registry/r-2579/Automated CSAM image scanning falsely flagged parents' medical photos of their children, Google, San Francisco and Houston, United States, 2021–20222026-10-10T16:06:44.084191+00:00In February 2021 Google's automated scanning flagged photos two fathers in San Francisco and Houston had taken of their young children's genital infections at medical professionals' request. Google reported both to authorities and disabled their accounts. Police in both cities cleared the fathers, but Google refused to restore one father's accounts or data even after police found no crime. The New York Times reported the cases. [United States, California; Texas]https://witnessedsentence.org/registry/r-2168/Facial recognition shoplifting watch-list falsely flagged customers, Rite Aid, United States, 2012–20202026-10-10T16:06:44.084191+00:00Rite Aid used facial recognition in hundreds of US stores, matching shoppers against a database of tens of thousands of often low-quality images of 'persons of interest'. False matches led staff to follow, search, expel or accuse customers, including an 11-year-old girl. Stores using the system were disproportionately in non-White areas. The FTC's proposed order banned Rite Aid from facial recognition surveillance for five years. [United States]https://witnessedsentence.org/registry/r-1976/Face recognition misidentification leads to wrongful arrest of Robert Williams, Detroit Police Department, Michigan, 2020–20242026-10-10T16:06:44.084191+00:00Detroit police ran a blurry surveillance still from a watch-store shoplifting through DataWorks Plus face recognition, which matched Robert Williams; he was arrested at home in January 2020 despite being out of state on the day of the theft. His 2021 federal lawsuit settled in June 2024, barring arrests based solely on face recognition results. [United States, Detroit, Michigan]https://witnessedsentence.org/registry/r-1029/Offender Assessment System (OASys) re-offending risk scoring, Ministry of Justice, England and Wales, 2001–present2026-10-10T15:46:29.971673+00:00The UK Ministry of Justice has used OASys, an algorithmic risk-assessment system, to score prisoners and people on probation for re-offending risk since 2001, completing over 1,300 assessments a day. The scores inform bail, sentencing and parole decisions. Academics, prisoners and a House of Lords committee have raised concerns over racial bias, inaccurate data and the inability to challenge errors. [United Kingdom, England and Wales]https://witnessedsentence.org/registry/r-0943/Police face recognition wrongful arrests, multiple US jurisdictions, 2020–20252026-10-10T15:46:29.971673+00:00Police in several US cities arrested people identified by face recognition searches without adequate follow-up investigation. At least eight people, most of them Black, were wrongly arrested. A 2025 Washington Post investigation found that officers often bypassed departmental rules on how face recognition results may be used. Detroit adopted new rules after a lawsuit. [United States, St. Louis, Missouri; Miami, Florida; Detroit, Michigan]https://witnessedsentence.org/registry/r-0769/"Catch and Revoke" AI social media surveillance program targeting international student visa holders, United States, 20252026-04-18T11:06:45.814679+00:00In 2025 US federal agencies used AI-assisted tools to screen the public social-media accounts of international students under a programme known as 'Catch and Revoke'. More than 1,600 student visas were revoked. Civil-society organisations, including EFF, argued that revocations targeted political speech and raised free-speech and due-process concerns. [United States]https://witnessedsentence.org/registry/r-0570/Abortion content moderation censorship, Meta (Facebook/Instagram/Threads), United States, 20252026-10-10T15:46:29.971673+00:00Meta's automated content moderation repeatedly removed abortion-related posts that did not break its own rules, including factual information about medication abortion. EFF's Stop Censoring Abortion campaign documented nearly 100 such removals in 2025, and accounts of clinics and researchers were suspended, with appeals often unanswered. [United States]https://witnessedsentence.org/registry/r-0524/Flock Safety ALPR abortion investigation, Johnson County Sheriff's Office, Texas, 2024–20252026-04-18T10:54:11.551330+00:00Johnson County Sheriff's Office in Texas queried Flock Safety's nationwide ALPR network of over 83,000 cameras to locate a woman who had self-managed an abortion, logging the reason as 'had an abortion.' Authorities framed it as a death investigation of a non-viable fetus. No charges were filed, but EFF-obtained documents contradict official denials that the search was unrelated to abortion enforcement. [United States, Texas]https://witnessedsentence.org/registry/r-0458/Gaggle surveillance software flags deleted draft email, leading to student suspension, Marana Unified School District, Arizona, 20252026-04-18T10:46:14.877322+00:00Gaggle surveillance software, deployed by Marana Unified School District in Arizona, flagged a never-sent draft email joke written by a student at home on a school-issued Chromebook. The student was suspended despite being off campus and outside school hours. EFF filed an amicus brief arguing the school violated First Amendment and due process rights. [United States, Arizona]https://witnessedsentence.org/registry/r-0149/Metropolitan Police live facial recognition: mass scanning and misidentification, London, 2016–20262026-10-10T15:46:29.971673+00:00The Metropolitan Police has deployed live facial recognition across London since 2016, scanning millions of faces a year. The force's own 2025 report found that 80% of people wrongly flagged were Black. After the system misidentified anti-knife-crime campaigner Shaun Thompson, he and Big Brother Watch brought a High Court challenge in January 2026, arguing the deployment lacks a legal basis and breaches privacy, expression and assembly rights. [United Kingdom, London]https://witnessedsentence.org/registry/r-0138/Facial recognition misidentification of ex-police officer as shoplifter, Facewatch system, Budgens, United Kingdom, 20262026-10-10T15:46:29.971673+00:00A former police officer was wrongly identified as a shoplifter by facial recognition technology provided by Facewatch and deployed by Budgens in the United Kingdom. She was asked to leave the store. Big Brother Watch criticised the lack of any appeals process and the opacity of the algorithmic flagging. The case was reported by the Financial Times in March 2026. [United Kingdom]https://witnessedsentence.org/registry/r-0132/Live facial recognition deployment by Essex Police, England, paused 2026 after accuracy and bias concerns affecting 2.5 million people2026-10-10T15:46:29.971673+00:00Essex Police deployed live facial recognition technology, scanning the faces of approximately 2.5 million people without adequately testing the system for accuracy or racial bias. Big Brother Watch and a Cambridge study raised concerns about ineffectiveness and discrimination. Essex Police paused the deployment in March 2026 following scrutiny of the system's bias and accuracy risks. [United Kingdom, Essex, England]https://witnessedsentence.org/registry/r-0099/RisCanvi risk-assessment algorithm, Catalonia prison system, Spain, 2010–present2026-10-10T15:46:29.971673+00:00Since 2010 Catalonia's prison system has used RisCanvi, a risk-assessment algorithm, to score inmates' risk; the scores inform decisions on leave, parole and prison benefits. A joint investigation by journalist Pablo Jiménez Arandia, Algorights and Público reported that the tool restricts vulnerable inmates' access to these benefits and lacks transparency and accountability. [Spain, Catalonia]